3.3.1 Lets compute descriptive statistics to help determine the distribution of numerical variables.

We can see the variable stunted_0_6_months has a large positive skewness. For modeling reasons we should consider a log or sqrt transformation on this variable to follow the normal distribution.

cema_clean %>% describe()

We can take the above analysis further by grouping the data by county and computing the descriptive statistics.

cema_clean %>% 
  group_by(county) %>% 
  describe() %>% 
  arrange(described_variables, desc(skewness))

3.3.2 Visualization of normality of numerical variables

We can see that all of the numeric variables in the dataset are positively skilled. We also see how the ditributions of these would look like when square root and log transformations are applied.

cema_clean %>% 
  plot_normality()

3.4.1 Calculation of correlation coefficient

The following is a correlation matrix show the association of the numeric variables in the dataset. All of them are positively correlated.

cema_clean %>% 
  correlate() %>% 
  plot()

## Analysis of Numeric variables over time

Lastly, as part of our exploratory analysis, we can leverage the period column to explore how the numeric indicators are changing over time by county. We can do this by plotting each numeric indicator against time, for each county. In the chunk below we filter for rows where county is Embu and plot the total dewormed children over time. The plot shows a relatively constant trend with conspicuous hikes in every May and November of every year.

ggplotly(cema_clean %>% 
  filter(county == "Embu County") %>% 
  ggplot(aes(period, total_dewormed))+
  geom_line()+
  geom_point()+
  labs(
    title = "Total Dewormed Children Over Time", x = "Date", y = "Total Dewormed"
  )+
  theme_bw()
)

Since we have several counties (47) in the data, visualizing all of them in one plot will be will create a visual hard to interpret. We can leverage some javascript to scale the faceting to all 47 counties and indicators. This will create an interactive display which one can search, sort, and filter for any indicator by county.

cema_clean %>% 
  mutate(county = str_remove(county, " County")) %>% 
  pivot_longer(
    3:last_col(), names_to = "indicator", values_to = "value"
  ) %>% 
  unite("analytic", county, indicator, sep = " - ") %>% 
  group_by(analytic) %>% 
  plot_time_series(
    period, value, 
    .interactive = F, .x_intercept = "Date", .y_lab = "Value",
    .title = "General Trend by County and Indicator", 
    .smooth = F
    )+
  geom_point()+
  # facet by all groups in the pivoted data
  facet_trelliscope(~ analytic, scales = "free", width = 1000, nrow = 3, ncol = 3, as_plotly = T, path = "./")

4.1.1 Interpretation

We see a correlation coefficient close to 1 (positive) in counties like Samburu, Busia, Kisii and Turkana. This suggests a strong positive association between Diarrhoea casess and the number of dewormed children in these counties. In other words, as the number of dewormed children increase in these counties, the number of diarrhea cases among children under 5 years will most likely increase.

We also observer a weak negative correlation coefficient in counties like Nyeri, Kakamega, Tharaka Nithi and Kitui. This suggests a weak negative association between Diarrhoea Cases and the number of dewormed children in these counties. This means as the number of dewormed children increase in these counties, the number of diarrhea cases among children under 5 years will most likely decrease.